Node Rewards & Slashing

InfraMind nodes earn rewards for successfully executing inference jobs and returning verifiable results. The protocol tracks each node's work output, latency behavior, reliability score, and stake exposure to determine eligibility, prioritization, and payment. The economic engine is designed to reflect real performance rather than static provisioning. Nodes are rewarded for what they execute—not what they promise.

The reward system is on-chain compatible and operates across epochs, with rewards distributed automatically or claimed via CLI depending on the operator’s configuration. All rewards are paid in the $INFRA token, the native settlement token of the InfraMind protocol.

Work Completed → Tokens Earned

Each successfully completed job results in a proof-of-serving receipt. This receipt is signed by the node and submitted either directly to the protocol reward oracle or batch-posted by the scheduler.

Example proof:

{
  "job_id": "c812-f3b7",
  "model_ref": "ipfs://QmXd123...",
  "latency_ms": 203,
  "node_id": "0xA9f12...",
  "output_hash": "0x7f8d...",
  "signature": "0x418abc...",
  "timestamp": 1720201314
}

Once the proof is verified, a reward is credited to the node’s address.

Automatic settlement (push model):

infra rewards auto-enable

Manual claim (pull model):

infra claim --job c812-f3b7

All receipts are checkpointed in a local Merkle tree, allowing later batch aggregation to reduce gas cost for reward redemption.

Reward Formula

Rewards per job are calculated as follows:

Reward = B × R × Q × (1 - P)

Where:

• B = base reward multiplier (based on job type)

• R = regional priority multiplier

• Q = quality coefficient (latency percentile, success ratio)

• P = protocol-defined penalty rate (if SLA drift is observed)

B: Job Type Weight

Job Type	Base Multiplier (B)
CPU Inference	1.0
GPU Inference	3.5
Session-Aware	2.2
Enclave (TEE)	4.8
zkML Verified	6.0

R: Regional Incentives

To encourage mesh diversity, some zones have reward boosts:

Region	Multiplier (R)
Africa-North	1.4
Asia-South	1.2
Europe-Central	1.0
US-East	0.9

Regions are inferred from INFRA_LOCATION_HINT and verified via latency tracebacks.

Q: Quality Coefficient

This dynamic multiplier is calculated from:

• Rolling latency percentile (faster → higher Q)

• Proof submission timeliness

• Historical job completion rate

• Response schema match

Example:

{
  "latency_ms": 194,
  "latency_percentile": 92,
  "success_ratio": 0.986,
  "timeliness": true
}

Yields:

Q = 1.0 + (0.92 × 0.5) + (0.986 × 0.3) = ~1.61

P: Penalty Factor

Nodes that exhibit SLA violations have a percentage deducted from their earned rewards.

Penalty types:

• Job decline after assignment → 5%

• Missed deadline → 10%

• Invalid proof format → 20%

• False resource declaration → Slash and blacklist

All penalties are published on-chain and persist through the current epoch.

SLA Enforcement and Slashing

Each node maintains a sliding-window SLA score:

SLA = (Jobs Completed) / (Jobs Assigned)

A node with an SLA below 0.85 enters probation. If it falls below 0.70 over the next 100 jobs, it is slashed.

Slashing logic:

• First violation: stake reduced by 15%

• Second violation: 30% + removal from scheduler index

• Third violation: temporary blacklist (14 days)

Slashed stake is distributed:

• 50% → protocol burn

• 30% → compensation pool for failed jobs

• 20% → redistribution to top-performing nodes in the same region

Slashing transaction:

infra slash --node 0xA9f12... --amount 34.2

Reward Distribution and Epochs

Rewards are finalized every 12-hour epoch. Nodes may configure delayed claim or streaming payout.

Streamed:

infra rewards stream --interval 15m

Batch claim:

infra rewards batch-claim

All payments are settled to the address associated with the node's public key.

Reputation Score

In addition to raw SLA, nodes maintain a public reputation score:

Reputation = 0.6(SLA) + 0.3(Job Quality) + 0.1(Uptime)

Nodes with high reputation get:

• Priority in job selection

• Access to private model pools

• Eligibility for zkML pilot workloads

• Higher reward multipliers (Q increases)

Reputation can be queried:

infra reputation

Or seen on the mesh explorer.

Example Reward Flow

1. Node receives job

2. Executes within 310ms

3. Submits valid proof

4. Scheduler verifies hash and latency

5. Oracle finalizes reward: Reward = 1.0 × 1.2 × 1.6 × (1 - 0.0) = 1.92 INFRA

6. Tokens credited to wallet

Conclusion

InfraMind rewards are precision-aligned with actual compute work. There is no inflationary minting, no idle rent, and no synthetic emissions. Nodes earn when they serve, and lose when they fail. This architecture enforces honest behavior, promotes regional diversity, and ensures that every token in the system represents verifiable execution that can be traced, audited, and proven.